Personalised Learning Environments based on Knowledge Graphs and

the Zone of Proximal Development

Yoshi Malaise

and Beat Signer

Web & Information Systems Engineering Lab, Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium

Keywords:

Personalised Learning, Knowledge Graph, Personalised Teaching Assistant Tool, e-Learning, Smart Learning.

Abstract:

The learning of new knowledge and skills often requires previous knowledge, which can lead to some frus-

tration if a teacher does not know a learner’s exact knowledge and skills and therefore confronts them with

exercises that are too difﬁcult to solve. We present a solution to address this issue when teaching techniques

and skills in the domain of table tennis, based on the concrete needs of trainers that we have investigated in a

survey. We present a conceptual model for the representation of knowledge graphs as well as the level at which

individual players already master parts of this knowledge graph. Our ﬁne-grained model enables the automatic

suggestion of optimal exercises in a player’s so-called zone of proximal development, and our domain-speciﬁc

application allows table tennis trainers to schedule their training sessions and exercises based on this infor-

mation. In an initial evaluation of the presented solution for personalised learning environments, we received

positive and promising feedback from trainers. We are currently investigating how our approach and concep-

tual model can be generalised to some more traditional educational settings and how the personalised learning

environment might be further improved based on the expressive concepts of the presented model.

1 INTRODUCTION

Learning the ropes of a new activity can often be a

daunting task, in particular if the learning material re-

quires a lot of prior knowledge. Situations like these

often lead to frustration since the learner has no way

to ﬁgure out the exact skills and knowledge they are

missing in order to properly advance on their path.

Likewise, for educators and coaches it can be difﬁ-

cult to provide proper guidance if they do not know

their students’ past learning trajectory and their cur-

rent proﬁciency levels.

In previous research, the use of knowledge graphs

has been proposed as a way to provide a semantic

representation of all relevant topics and their rela-

tions for a given domain (Rizun, 2019). In such a

knowledge graph the topics are represented as nodes

of a directed acyclic graph with the edges represent-

ing speciﬁc associations. These associations typically

indicate that a certain topic should be explored ﬁrst,

as the knowledge and skills learned in that topic are

necessary to master a more advanced topic. The for-

mal representation of topics can be used to provide

https://orcid.org/0000-0002-3228-6790

https://orcid.org/0000-0001-9916-0837

an overview of which topics a learner could study

next. Furthermore, by automatically navigating the

graph during diagnostic assessments, we can detect a

learner’s knowledge gaps. In more structured envi-

ronments with clear requirements to reach at interme-

diate milestones (e.g. the six year curriculum to obtain

a high school diploma is split into a different set of

learning objectives for each two-year interval), a pre-

deﬁned walk through the knowledge graph in the form

of a so-called learning path can be deﬁned. When stu-

dents move from one curriculum to another—for in-

stance after a relocation—the knowledge paths of the

two curricula can be compared to provide assistance

during the transition period (Ilkou and Signer, 2020).

We believe that there are great opportunities for

adaptive and personalised learning environments tak-

ing a user’s zone of proximal development into ac-

count. Our goal is to use knowledge graphs in com-

bination with a user’s acquired skills to automatically

detect and recommend the topics to be learned next in

the zone of proximal development. We present a tool

to assist Flemish table tennis trainers in teaching the

fundamental skills of their sport. It serves as a con-

crete personalised learning environment as well as a

proof of concept to explore how the approach could

be generalised to a broader learning context.

Malaise, Y. and Signer, B.

Personalised Lear ning Environments based on Knowledge Graphs and the Zone of Proximal Development.

DOI: 10.5220/0010998600003182

In Proceedings of the 14th International Conference on Computer Supported Education (CSEDU 2022) - Volume 1, pages 199-206

ISBN: 978-989-758-562-3; ISSN: 2184-5026

199

We start by highlighting some related work and

introducing the concept of proximal development.

Based on a literature study and a survey that we con-

ducted to ﬁnd out more about the needs of train-

ers teaching fundamental table tennis skills, we de-

rive a number of requirements for a personal learn-

ing environment solution. After presenting a proto-

type of such a personal learning environment based on

knowledge graphs and the zone of proximal develop-

ment, we discuss its generalisation to other domains

and provide some general conclusions.

2 BACKGROUND

Knowledge graphs have, for instance, been used to

add some advanced navigation to the Moodle

learn-

ing management system (Scherl et al., 2012), result-

ing in an increase in participants’ overall knowledge

compared to Moodle’s classical interface.

A key construct of Vygotski’s theory of learning

is the identiﬁcation of the zone of proximal develop-

ment (Vygotsky, 1978). The zone is deﬁned as the

space between what a learner can do unaided and the

tasks that are too difﬁcult to be performed even under

some guidance, as illustrated in Figure 1. Tasks in this

zone are optimal to improve a learner’s knowledge, as

they are challenging enough to push the learner, but

not impossible to master.

Tasks a

learner can

perform

unaided

Figure 1: Zone of proximal development.

We believe that enriching educational content

such as explanations, exercises and multimedia con-

tent with knowledge graphs can form the core of a

smart, adaptive and personalised learning environ-

ment, always suggesting exercises in a user’s zone

of proximal development. Thereby, less time is

“wasted” in performing assignments covering already

known topics and we can reduce the risk of demotivat-

ing learners and lowering their conﬁdence by present-

ing them too difﬁcult exercises (Shabani et al., 2010).

https://moodle.org

Similar research has recently been conducted

based on data made available via the Learnta TAD

online platform (Zou et al., 2019; Baker et al., 2020).

As a result of their evaluation of k-12 students in the

ﬁelds of Mathematics and English, the authors con-

cluded that in both ﬁelds the students managed to

complete more tasks if the tasks were selected from

their zone of proximal development. A major differ-

ence to our proposed solution is that they used statisti-

cal methods to estimate the mastery of topics whereas

we aim for a teacher-in-the loop approach.

In order to properly deﬁne the requirements for a

personalised learning tool for the ﬁeld of table tennis,

we relied on information provided to us by trainers in

a survey, as well as some literature research on the use

of knowledge graphs in education and models used

in sports education such as the three-stage model of

motor skill acquisition (Fitts and Posner, 1967) and

Kolb’s experiential learning cycle (Kolb, 2014).

Cognitive

Stage

Associative

Stage

Autonomous

Stage

Figure 2: Three-stage model of motor skill acquisition.

In the three-stage model of motor skill acquisition

shown in Figure 2, learners transition through three

distinct phases. In a ﬁrst cognitive stage, the learner

is confronted with a new skill and it takes a lot of

cognitive effort to perform basic actions. Once a user

gets familiar with a technique, they enter the associa-

tive stage and acquire how to adapt a technique based

on evolving or more difﬁcult situations. In the ﬁnal

autonomous stage, a motor skill can be performed al-

most completely from muscle memory, allowing the

learner to focus on the surrounding context.

Abstract

Conceptualisation

Active

Experimentation

Concrete

Experience

Reflective

Observation

Figure 3: Kolb’s experiential learning cycle.

Kolb’s experiential learning cycle theory—with

its four-stage cycle of learning illustrated in Fig-

ure 3—emphasises the importance of a learner’s ex-

periences while acquiring a new skill (Kolb, 2014).

In the active experimentation stage, the learner tries to

perform an action and experiences the result of their

action (success or failure) in the concrete experience

https://www.learnta.com

CSEDU 2022 - 14th International Conference on Computer Supported Education

200

stage. A learner then enters the reﬂective observation

stage where they reﬂect about their action and its re-

sult, potentially assisted by feedback from a supervi-

sor. The learner further moves to the abstract concep-

tualisation stage where they draw some conclusions

and plan how to adapt their action in a next iteration

through the cycle. In addition, Kolb also discusses

four learning styles, each of them related to different

stages of the cycle. This theory has later been ex-

tended by Honey and Mumford (Honey, 1992), stat-

ing that learners can be placed along the two axes

highlighted in Figure 4.

Perception Continuum

Processing Continuum

Pragmatists Theorists

ReflectorsActivists

Concrete

Experience

(Feeling)

Abstract

Conceptualisation

(Thinking)

Reflective

Observation

(Watching)

Active

Experimentation

(Doing)

Figure 4: Learning Styles by Honey and Mumford.

The perception continuum differentiates between

people who prefer thinking and reasoning about the

theory of new situations before partaking, and peo-

ple who prefer to gain an intuition about what to do

in a certain context. The processing continuum dis-

tinguishes between those who prefer to learn from

observing others ﬁrst and those who prefer to ﬁgure

things out as they go. Based on these axes, four dis-

tinct learning styles are identiﬁed, each of them learn-

ing best during a different stage of the learning cycle.

By ensuring that we cycle through all four stages, we

guarantee that any of these four types of learners have

a chance to grow their knowledge.

3 SURVEY

In order to learn more about the needs of table tennis

trainers, we conducted a survey with trainers regis-

tered at the Vlaamse Tafeltennisliga (VTTL)

and the

table tennis division of Sporta

, the two largest Flem-

ish table tennis organisations. From the invited train-

ers, 82 participated and ﬁlled in the survey. A major

https://www.vttl.be

https://www.sportateam.be/tafeltennis

part of the responses (around 40%) came from train-

ers with more than ten years of experience, and we got

a healthy mix of both, experienced trainers leveraging

years of expertise as well as less experienced trainers

looking forward to introducing new ideas.

Our survey started with a short introduction about

the planned research. This was followed by a number

of questions divided into distinct sections, including

the trainers’ experience, the material they use during

training sessions, the way they analyse players, how

they communicate with players, their preparations of

training sessions as well as general feedback. The de-

tailed questions that were used in our survey as well

as the anonymised results are available online

. Based

on our literature study and the results of our survey,

we derived the following eleven core requirements

for a technology-enhanced personalised learning en-

vironment for the domain of table tennis:

R1: Trainers Should Be Able to Manage Exercises.

From the open questions of our survey, it became

clear that trainers would like to the have the freedom

to experiment with their own exercises and content.

This indicates that trainers should be able to inspect as

well as add new exercises to a learning environment.

For every exercise it should be possible to manage its

name, a description and image, as well as the tech-

niques it teaches and any prerequisites in the form of

techniques a user already must know.

R2: Trainers Should Be Able to Manage the Evalu-

ation Criteria for a Technique. The evaluation cri-

terion forms the cornerstone in the feedback process

to players (one of the important aspects of Kolb’s ex-

periential learning cycle) and we should strive to im-

prove upon the criteria as well as offer trainers the

ﬂexibility to emphasise certain points. In order to

achieve this goal, trainers should be able to inspect all

the techniques and the way they are related. Further,

it should be possible to inspect a technique’s evalua-

tion criteria as well as to add or delete an evaluation

criterion.

R3: Trainers Should Be Able to Manage Assess-

ments. Assessments are used to validate whether a

player masters a certain technique; a necessary fea-

ture if we want our solution to be able to offer only

those exercises that are in the zone of proximal devel-

opment. To ensure that trainers do not feel forced into

one way of doing things—which was a major point of

feedback in our survey’s open question—assessments

should be manageable by trainers. They should be

able to add new assessments, declare the outcomes

of passing and assessment and select the correspond-

https://doi.org/10.5281/zenodo.6091625

Personalised Learning Environments based on Knowledge Graphs and the Zone of Proximal Development

201

ing exercises (together with the minimal requirements

that a player should meet) for an assessment.

R4: Trainers Should Be Able to Supervise Assess-

ments. When a player is performing an assessment,

a trainer should be able to grade the player’s perfor-

mance. Based on this grading, the system should be

able to update a player’s proﬁle. The reported per-

formance can be used in future communication with

a player and improve the communication process, a

demand that was indicated by half of the trainers in

our survey. Further, the updated user proﬁle allows

the system to later ﬁlter out exercises that are either

too difﬁcult or too easy.

R5: Trainers Should Be Able to Manage the Player

Knowledge Graph. It should be possible for a

trainer to inspect a player’s current knowledge, in-

cluding their proﬁciency level for individual tech-

niques. Further, a trainer should be able to manu-

ally update the data about a player’s knowledge. This

functional requirement enables trainers to ﬁne tune

the model in case something went wrong, or to boot-

strap the process for newly joining experienced play-

ers.

R6: Trainers Should Be Able to Inspect a player’s

Past Performance. In order to provide proper long-

term supervision, trainers need to have a good men-

tal model of the strengths and weaknesses of their

players. However, 68% of the survey participants in-

dicated that they do not keep any notes about their

players and solely rely on their memory. To better

assist trainers, they should be able to inspect the re-

sults of a player’s past assessments and training ses-

sions, including the grading of speciﬁc evaluation cri-

teria. This is particularly useful in clubs where multi-

ple trainers are teaching the same players.

R7: Trainers Should Be Able to Prepare Train-

ing Sessions. It should be possible for a trainer to

prepare a session with minimal effort, given that in

our survey a lack of time was the most common

reason why trainers do not prepare sessions. This

preparation should include the selection of exercises

from the list of recommended exercises in a player’s

zone of proximal development (automatically gener-

ated based on their current proﬁciency level for dif-

ferent techniques). A trainer should further be able to

select certain exercises based on a speciﬁc technique.

R8: Trainers Should Be Able to Supervise Training

Sessions. While supervising sessions, a tool should

ﬁrst show the exercises to players before they start

performing them. Once the speciﬁed time for an ex-

ercise has elapsed, this should be indicated to the

trainer. A trainer should then have the possibility to

grade the player’s performance for each of the evalu-

ation criteria of the techniques forming part of the ex-

ercise. This also provides an ideal opportunity for the

trainer to immediately provide some feedback (based

on the evaluation criteria questions) to the player who

can then reﬂect before starting a next cycle as pro-

posed in Kolb’s experiential learning cycle.

R9: The Application Should Adapt to Different

Screen Sizes. The use of large and heavy devices,

such as laptops, can be cumbersome during training

sessions. More than half of the survey participants

indicated to use their smartphone during a session,

while less than 10% use a tablet or laptop. Therefore,

an application should adapt to arbitrary screen sizes.

R10: The Application Should Allow the Sharing of

Data between Trainers of a Club. Currently, one

of the main challenges is that the sharing of informa-

tion between trainers is an ad-hoc process with less

than 20% of the survey participants having a system

to share information. An application should therefore

enable data sharing between trainers of the same club.

R11: Data of a Club Should Be Private to That

Club The training progress and contact details of

members is sensitive information. It should therefore

not be possible for members of a different club to gain

access to this information.

4 LEARNING ENVIRONMENT

Our solution for a personalised learning environment

consists of two major parts. First, there is the general

conceptual model and framework to represent all the

data and metadata necessary for the general knowl-

edge graph as well as a player’s expertise about parts

of this graph (player knowledge graph). Second, we

have developed a custom progressive web application

tailored towards the needs of trainers (Malaise, 2021).

Selectors Resources

Entities

Links

source target

Layers

Users

Context

Properties

Figure 5: Overview RSL hypermedia metamodel.

In order to model the complex domain-speciﬁc

knowledge, we opted to use the resource-selector-

link (RSL) hypermedia metamodel (Signer and Nor-

rie, 2007) shown in Figure 5. A Resource is repre-

senting any real or virtual object, such as an image, a

CSEDU 2022 - 14th International Conference on Computer Supported Education

202

Figure 6: Conceptual model.

student, a speciﬁc skill level or an arbitrary document.

Often one does not want to refer to an entire resource,

but to address parts of it (e.g. a speciﬁc person in an

image) via the concept of a Selector. Finally, a Link

can be used to represent a bidirectional relationship

between two or more entities.

These three core types are all subtypes of the more

general Entity type and as such can have their own

properties. This further implies that a link can not

only be used to deﬁne relationships between resources

but also have selectors or even other links as source

or target. It is out of the scope of this paper to de-

scribe the full RSL metamodel but further details can

be found in (Signer and Norrie, 2007).

The conceptual model of our personalised learn-

ing environment for the domain of table tennis is

shown in Figure 6. The rectangles represent spe-

ciﬁc RSL Resources and the directed arrows repre-

sent RSL Links between Entities. Further, when a link

itself is used as the source of another link, this is indi-

cated via a black solid circle on the source link.

A Player represents a person who is a member of a

club for which a trainer is creating training schedules.

The goal of our solution is to assist trainers in improv-

ing the skills of players and therefore players form

a central component of our conceptual data model.

A Technique is a certain skill or a piece of knowl-

edge that the player needs to master in their path to

proﬁciency in a sport discipline. A technique might

require a certain level of proﬁciency of other tech-

niques before it can be acquired, which is represented

by the Requires link and the MinimalLevel link over

this Requires link. For example, a technique could be

the act of hitting the ball. Before a player can serve,

they should be able to hit the ball. This implies that if

a player is having trouble serving, it could be caused

by the fact that they have not yet sufﬁciently mastered

the skill of hitting the ball. The ProﬁciencyLevel is the

level at which a player has mastered a speciﬁc skill or

technique and is based on the three-stage model of

motor skill acquisition introduced in Section 2.

The EvaluationCriterion consists of a basic ques-

tion about parts of a technique (e.g. “Did the player

place their feet the correct way?”) and is associated

with the corresponding technique via the ValidatedBy

link. These questions can be used to evaluate a

player’s performance (score between 1 and 5) after

each exercise. Further, a trainer might discuss the

criteria with the player and provide immediate feed-

back, pushing the player to reﬂect and improve. An

EvaluationCriterionResult contains information how

well the player performed for a speciﬁc evaluation

criterion. The results for a given evaluation criterion

are linked to the exercise performed when the trainer

graded the criterion via the PerformedExercise link.

This contextual information is useful since when a

player meets certain requirements of some exercises

but others not, it might indicate underlying issues with

other techniques used in the exercises.

An Exercise describes a situation in which a

player can learn about or improve on one or multi-

ple techniques as indicated by the Teaches link. Via

the Requires link it is possible to indicate that an ex-

ercise depends on certain techniques. Before a player

Personalised Learning Environments based on Knowledge Graphs and the Zone of Proximal Development

203

is able to perform the exercise, they need to master

a technique at least on the proﬁciency level indicated

by the MinimalLevel link deﬁned over the association

between an exercise and a technique. This is neces-

sary to ﬁlter out those exercises that are currently too

difﬁcult for the learner, in order to stay within their

zone of proximal development.

An Assessment is a special set of exercises re-

ferred to via a HasExercise link and designed to

gauge the proﬁciency of a player in certain tech-

niques. The HasExercise link further forms the

source of the MinimallyRequires link that points to

an AssessmentRequirement stating the objective and

measurable requirements for a speciﬁc exercise in the

assessment. It consists of a certain condition that

needs to be fulﬁlled for an exercise in order to say

that the player mastered the technique at a sufﬁciently

high level. For example, in an assessment on the abil-

ity of players to perform the forehand stroke, there

could be an exercise stating that they would need to

hit a piece of paper placed on the table with the ball.

The requirement could state that within 5 minutes a

player should place at least 20 strokes in the indi-

cated area. An AssessmentResult manages informa-

tion on how a player performed for a given assessment

by linking to the corresponding RequirementResults.

When an assessment has been passed successfully, it

will lead to the creation of new Masters links asso-

ciating a player with the techniques they now master

at a speciﬁc level. The level of mastery that has been

achieved is represented via the AtLevel link deﬁned

over the Masters link. The LearningOutome that is re-

lated to an easement via the Awards link deﬁnes which

Masters and AtLevel links will be created. Based on

a player’s achieved proﬁciency levels, our application

is able to suggest or hide certain exercises.

A Training deﬁnes the training schedule for an

individual session on a speciﬁc day, including all

the exercises that need to be performed as indicated

via the IncludesExercise link (note that this link will

normally point to multiple targets representing the

exercises). Information about how the player per-

formed during a training is further managed in a

TrainingResult. A training is linked to its result via a

HasResult link and training results can provide valu-

able insights on how a player is evolving and hope-

fully improving over multiple training sessions.

Aside from general comments about a player’s

performance during a training, a TrainingResult nor-

mally consists of multiple EvaluationCriterionResults

as represented via the Includes link. Thereby, an

EvaluationCriterionResult holds the score given by the

trainer on how well the player scored on a certain

EvaluationCriterion (referred to via the BasedOn link)

while performing a speciﬁc Exercise (indicated by the

PerformedExercise link). This ﬁne-grained model al-

lows us to look for speciﬁc patterns or underlying rea-

sons why players score well for an EvaluationCriterion

in one exercise while they might perform poorly for

the same criterion in another exercise.

While the conceptual data model forms the core of

our application, the manual manipulation of the graph

via queries would lead to a poor user experience and

be too technical for the average trainer. Therefore, we

developed a server based on the Spring Boot

frame-

work, providing endpoints to request all important de-

rived results as well as to register new information. To

interact with the server, a progressive web application

that can either be used as a desktop application or on a

mobile device has been realised. The web application

was developed using the angular framework

. The ap-

plication allows trainers to design personalised train-

ing schedules for players and supports the supervision

of assessments without requiring a user to know any-

thing about the underlying knowledge graph.

5 USE CASE

We discuss some of the most important aspects of the

user facing application as well as how a trainer is go-

ing to interact with them

. In order to make sure that

a trainer can utilise their own expertise and creativ-

ity, they have to be able to add their exercises to the

system (requirement R1). When doing so, they can

specify a name, a duration, an image, any number of

tags and a description. They can indicate which tech-

niques need to be mastered at which proﬁciency level

before a player can perform the exercise, and the skills

taught by the exercise. This metadata is going to be

used to suggest exercises to individual players.

The knowledge representation of a speciﬁc

player consists of a detailed representation of their

achieved proﬁciencies and knowledge (player knowl-

edge graph). The direct manipulation of this graph

would be too technical and we provide trainers the

possibility to create an assessment (requirement R3),

where they can specify which skills they would like to

assess as well as the proﬁciency level. The application

then suggests exercises that are a good ﬁt for the as-

sessment. For every selected exercise, the trainer has

to provide a minimal requirement condition needed

for a player to pass that part of the assessment, which

is later used to modify the player knowledge graph.

https://spring.io/projects/spring-boot

https://angular.io

https://youtu.be/OSw2PWpG6dg

CSEDU 2022 - 14th International Conference on Computer Supported Education

204

When a trainer feels that a player is ready, they

can ask them to take part in an assessment (require-

ment R4). Each exercise is shown with a built-in

countdown timer and after time is up, a dialogue asks

the trainer whether the player passed the requirement

for the exercise. If a player has passed all the re-

quirements at the end of the assessment, they will

be awarded the conﬁgured proﬁciencies. In case they

failed some exercises, the system will check whether

the failed exercises contained some dependencies to

skills that did not show up in the dependency list of

the exercises the player passed. If such a skill ex-

ists, it might likely form a knowledge/skill gap and

the system will report this information to the player.

When planning a session for a speciﬁc player,

our solution will automatically ﬁlter out all exercises

that are either too difﬁcult or too easy for the player

based on the proﬁciency data stored in their knowl-

edge model. This guarantees that all exercises se-

lected by the trainer will be situated in the player’s

zone of proximal development. Our application fur-

ther allows trainers to search for speciﬁc exercises

based on a name, tags or speciﬁc techniques that they

would like the player to work on (requirement R7).

When supervising a training session, the trainer is

shown a screen with the image, name and description

of the exercise as well as some controls to operate the

system. Once the time for an individual exercise has

elapsed, a list of evaluation criteria for the techniques

taught by the exercise, as well as the techniques di-

rectly required by these techniques, are shown to the

trainer for detecting errors in prior techniques as well

as knowledge and skill gaps. Further, the trainer

can immediately provide this detailed feedback to a

player, supporting the reﬂective observation and ab-

stract conceptualisation phase of Kolb’s experiential

learning cycle (requirement R8). Once the training

session is completed, a trainer can add more general

feedback for future analysis (requirement R6).

Our solution contains some dedicated views to

inspect the progress of individual players (require-

ment R6), including a spider chart of the most recent

average evaluation criteria scores of techniques they

have been working on, a graph showing the evolu-

tion of their scores for certain techniques over time, as

well as an enhanced knowledge graph view in which

a trainer can see and modify the achieved proﬁciency

of each technique (requirement R5) in the form of a

player knowledge graph as illustrated in Figure 7.

While our solution is well suited for one-to-one

sessions, various trainers asked us to also include

some functionality to support group sessions. In or-

der to address the needs of these trainers, we decided

to also add a group mode where an exercise will only

Figure 7: Player knowledge graph.

be available if each group member satisﬁes the nec-

essary requirements. Further, there is a single eval-

uation window, containing a table with all the tech-

niques used throughout the training on one axis and

all participating players on the other axis, where a

trainer can report a player’s performance for a spe-

ciﬁc technique (requirement R8).

6 EVALUATION

The feedback from an experienced group of trainers is

essential to ensure that our application with its knowl-

edge graph-based approach will not only be useful in

real life, but that trainers would also like to use the ap-

plication. While various table tennis clubs indicated

to be available and willing to participate in trial runs,

it was unfortunately not yet possible to perform an in

situ evaluation due to the global pandemic after the

Covid-19 outbreak. The best alternative was the cre-

ation of a video clip with some voice-over describing

the application and illustrating how the tool could be

used by a trainer, both during the preparation and the

supervision of a training session. The video was then

emailed to all the trainers who indicated to being open

for further questions in our initial survey, and we re-

ceived a total of 15 responses from trainers ﬁlling in

the second survey. Everybody ﬁlling out the survey

was invited for a 30 minute video interview to discuss

our solution and provide further feedback, and four

trainers accepted to participate in an interview.

When asked whether they would like to use our

solution in their club, three out of the 15 participants

answered with a neutral score of 3 on a 5-point Lik-

ert scale, six trainers answered with a score of 4 and

another six participants answered with a score of 5.

These scores indicate that most trainers seem to be

excited about the idea of using the tool and the results

are quite promising. We noticed some recurring fea-

ture requests such as providing an ofﬂine ﬁrst version

and improving the mobile user experience.

Personalised Learning Environments based on Knowledge Graphs and the Zone of Proximal Development

205

The trainers who took part in the interview all

showed enthusiasm and were excited to use the sys-

tem in a real environment. A main topic in most inter-

views was the question of who should be able to mod-

ify the general knowledge graph, as well as whether

we should push for exercises created in one club to

automatically be available to all clubs; an interesting

topic that will need some further investigation.

7 DISCUSSION

We presented a prototype of a personalised learning

environment helping table tennis trainers in preparing

tailor-made training sessions. Our conceptual model

is not limited to table tennis, but could already be used

as is in tools for most individual sports disciplines.

Further, our conceptual model might be generalised to

support more traditional educational settings as well

as hybrid classroom setups where on-site classes are

combined with partial self-study trough e-learning.

For instance, the three-stage model of motor skill

acquisition might be replaced with Bloom’s taxon-

omy (Bloom, 1956). Further, assessments and the

analysis of skill gaps should translate well to remote

e-learning environments and private tutoring sessions.

In addition to suggesting the right content for

learners, our model could also be used to efﬁciently

analyse the current knowledge levels of newcomers.

A student might be given a set of exercises about a

speciﬁc skill and if they fail to perform the exercise

adequately, we can suggest exercises based on the di-

rect requirements of that skill. If a student manages to

complete that exercise, it implies that the prerequisite

skill is not the issue; otherwise we need to analyse

where the knowledge gap causing the student to fail

that prerequisite is coming from.

Past research has shown that students perform bet-

ter when the material has been adapted to their pre-

ferred learning styles (Mustafa and Sharif, 2011). A

main advantage of our model being based on the

RSL hypermedia metamodel is that it allows us to take

personalisation even further. Instead of simply sug-

gesting different exercises based on a learner’s pro-

ﬁciency, we might adapt the exercises based on the

RSL model’s concept of structural links and their use

for adaptive document structures (Signer, 2010).

8 CONCLUSION

We have presented a prototype of a technology-

enhanced personalised learning environment for the

domain of table tennis, making use of knowledge

graphs in combination with the results of assessments

to suggest exercises in a player’s zone of proximal

development. The discussed research on a concep-

tual model and domain-speciﬁc application represents

a step towards a personalised learning environment

where the learner is central. We do not only aim to

provide the right content at the right time, but also en-

vision to further adapt the presented content based on

the underlying RSL hypermedia metamodel in com-

bination with a learner’s preferences, their previous

experience as well as their learning style.

REFERENCES

Baker, R. et al. (2020). The Results of Implementing Zone

of Proximal Development on Learning Outcomes. In

Proc. of EDM 2020.

Bloom, B. S. (1956). Taxonomy of Educational Objectives.

Vol. 1: Cognitive Domain. New York: McKay, 20(24).

Fitts, P. M. and Posner, M. I. (1967). Human Performance.

Brooks/Cole Publishing Company.

Honey, P. (1992). The Manual of Learning Styles. Peter

Honey Publications, third edition.

Ilkou, E. and Signer, B. (2020). A Technology-enhanced

Smart Learning Environment Based on the Combina-

tion of Knowledge Graphs and Learning Paths. In

Proc. of CSEDU 2020.

Kolb, D. (2014). Experiential Learning: Experience as the

Source of Learning and Development. Pearson Edu-

cation, second edition.

Malaise, Y. (2021). iSport: Towards a Smart Trainer Assis-

tant. Master’s thesis, Vrije Universiteit Brussel.

Mustafa, Y. E. A. and Sharif, S. M. (2011). An Approach

to Adaptive E-Learning Hypermedia System Based on

Learning. International Journal of Library and Infor-

mation Science, 3(1).

Rizun, M. (2019). Knowledge Graph Application in Educa-

tion: A Literature Review. Acta Universitatis Lodzien-

sis: Folia Oeconomica, 3(342).

Scherl, A. et al. (2012). Interactive Knowledge Net-

works for Interdisciplinary Course Navigation Within

Moodle. Advances in Physiology Education, 36(4).

Shabani, K., Khatib, M., and Ebadi, S. (2010). Vygotsky’s

Zone of Proximal Development: Instructional Impli-

cations and Teachers’ Professional Development. En-

glish Language Teaching, 3(4).

Signer, B. (2010). What is Wrong with Digital Documents?

A Conceptual Model for Structural Cross-Media Con-

tent Composition and Reuse. In Proc. of ER 2019.

Signer, B. and Norrie, M. C. (2007). As We May Link:

A General Metamodel for Hypermedia Systems. In

Proc. of ER 2007.

Vygotsky, L. (1978). Mind in Society: Development of

Higher Psychological Processes. Harvard University

Press.

Zou, X., Ma, W., Ma, Z., and Baker, R. S. (2019). To-

wards Helping Teachers Select Optimal Content for

Students. In Proc. of AIED 2019.

CSEDU 2022 - 14th International Conference on Computer Supported Education

206